Purpose
Interest in foveal pit shape has grown due to its possible links to race, gender, refraction, amblyopia and retinal pathology and the increasing availability of high-quality OCT scans. Foveal pits with flat bottoms and/or asymmetric walls were poorly fitted using the Difference of Gaussians (DoG) models in previous studies. We explored more sophisticated mathematical models to improve goodness of fit and to capture important anatomic characteristics.
Methods
Raw data from 3501 EMM5 foveal scans (RTVue 100) from 582 young adults (421 myopes and 161 matched non-myopes, mean age=21.9±1.4 yrs) were used to test fitting of the DoG model and a Sloped Piece Gaussian (SPG) model. Three consecutive scans were obtained from 99.4% of the eyes. The SPG was a linear combination of a tilted line and a piecemeal Gaussian function (two halves of a Gaussian connected by a straight line) with 6 parameters (the σ of the Gaussian, the center (C) and width (P) of a flat bottom, the slope of the tilted line (T), the height and vertical position of the curve. See figure). The 803-pixel (6 mm) horizontal and vertical retinal thickness curves (ILM to RPE) passing through the foveal center were first smoothed using an 80-pixel span to determine the first local maxima on both sides of the pit. The SPG model was then fitted to the pit data between the two maxima. Data with 50 additional pixels beyond the two maxima were fitted by the DoG. A nonlinear least squares method was used to find the best fitting curves. SPG and DoG were compared using t-tests.
Results
The SPG model produced significantly smaller root mean square errors than the DoG (3.9 vs. 7.1 μm, t=88.8, p<0.0001, vertical; 4.2 vs. 10.9 μm, t=116.3, p<0.0001, horizontal). The SPG revealed a consistent tilt in >95% of the curves with the nasal side higher than the temporal side by 11.5±0.70 μm (t>32.9, p<0.0001), similar to the 13 μm asymmetry from a histologic study (Curcio et al, IOVS, 2011). Such asymmetry was much less pronounced in vertical scans and might be related to the papillomacular bundle. The pit bottom width ranged from 0 to 600 μm, with 40% and 54% of the pits having a flat bottom >30 μm in horizontal and vertical directions, respectively. None of these characteristics was captured by the DoG model.
Conclusions
A SPG model provided superior fits to OCT foveal scans and captured a wider range of foveal pit anatomical characteristics than previous models.